FinOps in Action

FIA - Pathik Sharma
===

Pathik Sharma: [00:00:00] everything is so connected with each other, right? you, cannot just purely focus on, let's say, optimizing cost without thinking about performance and security and system design and scalability and stock outs and capacity and all of that.

And I think with that, the proliferation of FinOps and FinOps adjacent teams have gotten more and more tightly coupled with each other.

Intro: welcome to FinOps in Action. I'm your host, Taylor Houck. Each week I'll sit down with FinOps experts to explore the toughest challenges between FinOps and engineering. This show is brought to you by 0.5, empowering teams to optimize cloud costs with deep detection and remediation tools that actually drive action.

Taylor Houck: Hello, and welcome to another episode of FinOps in Action. guest sits in one of the more unique seats in FinOps. is the cloud cost optimization lead at Google Cloud, where he has helped [00:01:00] co-found Google's Cloud FinOps practice now leads cost optimization work with some of the largest enterprises in the world. This gives him a rare vantage point. He gets to see what one of the world's most AI-forward cloud providers is building from the inside and also work directly with customers who are trying to make cloud and AI economics work in practice. has led more than 100 customer optimization workshops, contributed to Google Cloud's cost optimization architecture guidance, has written extensively on FinOps and spoken at events like FinOps X and Google Cloud Next, and has helped shape FinOps and cost optimization strategies for many major enterprises. very interestingly for today's conversation, he is leading Google's internal work on agentic FinOps. to the show, Pathik Sharma

Pathik Sharma: Love the energy, Taylor. Thank you for the introduction. Um, [00:02:00] um, as you, as you mentioned, right, I think what we want to do and our team's mission is to, um, help customer build their future with AI, uh, without them accidentally, uh, getting bankrupt in, in the process of doing so. and our team comes into the play where are you using the right tools?

Are you making the right decisions? And, you know, thinking about business value, uh, and cost optimization hand in hand, so that, you know, everyone is, you know, nobody's getting bill shocks at the end of the month, right? We want to be more proactive. We want to be more governed and, uh, thinking about it more from a governance standpoint.

Um, and, uh, as you've mentioned, right, like that's something that we find that with AI, this is really something that is on top of many people's mind.

Taylor Houck: Yeah, it's something that, you know, it's been important for the past several years, and we've seen people be increasingly and increasingly interested in these types of topics. But now when you fold [00:03:00] in the acceleration of AI, especially in, let's call it like the fourth quarter of '25 into the first half of '26, it's becoming even more important

Pathik Sharma: it's funny you say this. So two years ago I was having this conversation about, uh, AI and FinOps and the intersection of this, and this was before, uh, it was coined AI for FinOps and FinOps for AI. And the conversation was sort of brushed off. It's like, "What are you talking about?

We're still figuring out cloud and, you know, for AI we are like way off," right? This was like three years ago. Uh, and now when we look at the survey data from FinOps Foundation, AI for FinOps, FinOps for AI, like these two sits right there, two things out of top three priorities that everyone is talking about

Taylor Houck: I think it just comes down to the acceleration of the capabilities of these tools,

Pathik Sharma: Yeah

Taylor Houck: back a couple years ago, it was fun. Some companies were finding valuable use cases, [00:04:00] but it was more so just like a very small sliver of the overall cloud footprint.

Pathik Sharma: Yes

Taylor Houck: But now we're seeing it accelerate so rapidly, even if it's still, let's call it 5%, 10%, whatever it is of your total cloud spend, if you look at the growth rate and the acceleration, you're realizing, holy cow, could end up being, you know, a material portion of my total cloud spend, potentially even surpassing my total cloud spend if

Pathik Sharma: Yes

Taylor Houck: over, you know, the next coming quarters or even years

Pathik Sharma: Yes, yes. I, I think there was this graph shared at some point, which is like the pace of technology overall, uh, what it could do in terms of decades, cloud was able to accelerate in, in, in terms of, uh, you know, number of years, like 10 years, 20 years, it was able to catch up with what technology has done for like last 100 years.

And then AI is now catching up in one or two years that cloud has done in decades. [00:05:00] So i- it's, uh, it, it's the pace as you've mentioned, right? And I think with that comes so many change, right? New products, new features, how your teams are using AI. Are you using AI or not? If you're not using AI, are, are we ho- holding ourselves back?

There is, there is this race in the industry that no matter what it is, you, you, you have to use AI. And I think to your point, two years ago, people are tinkering with it, uh, it was a fun idea, it was an experimentation thing. Now we have enterprise customers who are leveraging AI and that is impacting their top line, bottom line.

Uh, it has become a larger portion of their P&L. So I think, I think things are getting serious much more quickly than ever before. And I feel like this is this flywheel, right? Like 20 2011, 20 2012, everyone is thinking about, "Oh, we need to think about governance and cost management for cloud." And now it's all about, "Oh, we need to do it for [00:06:00] AI," right?

That becomes your new filling the blank.

Taylor Houck: Yeah, and there are so many different topics we could get into around this, and I really want to get into so much of it with you because of the, the-- where you sit at Google and also working with so many enterprises in the field. But kind of at a high level, you already touched on it. You have the FinOps for AI and the AI for FinOps.

I want to start broad. There's this term that I've been hearing more and more recently of agentic FinOps. where you're sitting inside Google, how are you thinking about agentic FinOps? What does this mean to you?

Pathik Sharma: Yeah. So I, I think like in working with customers for last seven, eight years, especially in the field of FinOps, it's-- we, we come down to culture, right? Culture is going to be the most important thing. We-- There is going to be a proliferation of tools that are out there, uh, in terms of like, hey, identifying re-recommendations and helping me optimize them, help me educate on new products and features and so on and so forth.

I think one of the gaps that [00:07:00] customers still have is knowing and doing, right? How do we fill that gap? And, and, and that's essentially where we think about agentic FinOps. And this is the first time in the history where a technology, specifically AI, is now being thought about as a human augmentation rather than just a piece of code that can do XYZ better, faster, more efficient, right?

So, so in, in, in-- To your point, I think what-- W-Working with Google definitely gives me the opportunity to work with some of the fastest growing companies that are out there, some of the longest standing companies that are out there, right, hundreds of years, uh, cloud native who have not even seen the data center and what-- how it operates.

Um, so sort of all s- form [00:08:00] and size and shapes of customers and AI is really the topic of conversation for all of these. And when we have this conversation about, okay, how do we think about driving FinOps forward? Um, it almost always comes down to, do we have the right data? Um, how do we understand the pricing, the billing, different services, the consumption model, the resources that we are using?

Um, how do we bridge the gap between knowing something and taking an action? And I think this is really where agentic framework for FinOps is crucial because now it takes you know, you understanding, doing the analysis, validating it, and then taking action. The whole thing end-to-end can now be done through an army of agents, right?

That's how we think about it. And to give you a simple example of this is, [00:09:00] um, let's say Kubernetes, tends to be one of the notorious things to optimize, uh, because there is the platform team who manages the, the entire platform, and then there is the application team who just wants to build products and ship features, and they don't care about the underlying infrastructure.

Um, and then there is a FinOps team, then there is a SRE team, and all needs to work together if they want to run a lean Kubernetes deployment at scale, right? So I'm talking hundreds of thousands of containers or maybe pods, even clusters and nodes and things of those nature. So in this case, AI can actually understand your goal.

My goal is to run a lean and efficient Kubernetes deployment that helps, let's say, for retail customers, a better checking out experience, a better search and discovery within my website, uh, a better payment profile, and so on and so [00:10:00] forth, right? So you- your end goal is still, hey, we want customers to have a delightful experience on our platform, but how do we not do that without breaking our bank?

And I think that's where this t- this goal can be disintegrated into several tasks. And specifically talking about FinOps, that is one agent that does analysis, understands what your cost and usage has been, utilization, request used versus, uh, requested versus used, and so on and so forth. Finding out and, uh, uh, uh, recommendations, then cross-verifying it with corporate policies, then checking and make sure that, hey, whatever implementation that we are recommending, does that make sense?

Whether I'll be doing it for production versus non-production, and then going in and actually implementing it, right? So it can actually take that goal, disintegrate into task, and, uh, do it all end to end. Like, that's how we are thinking about agentic FinOps. We are barely scratching the surface, but I think future is bright.

Taylor Houck: Patrik, is it [00:11:00] FinOps? Like, do you call this FinOps? Because what you had just described, using AI across this entire use case of how to run an efficient, you know, Kubernetes deployment, you're touching on a lot of different aspects of kind of cloud operations,

Pathik Sharma: Yes.

Taylor Houck: DevOps

Pathik Sharma: Yes.

Taylor Houck: SRE.

Pathik Sharma: Yes

Taylor Houck: it squarely FinOps?

Or how do you think about how this entire industry is going to evolve?

Pathik Sharma: Yeah. I think this is at, at the end of the day, right, like if we go back to is it the FinOps team doing FinOps versus FinOps team's job to empower others to do FinOps, right? And I think that's where we are, uh, going to get into this agentic era, is everything is so connected with each other, right? you, cannot just purely focus on, let's say, optimizing cost without thinking about performance and security and system design and scalability and stock outs and capacity and all of that.

And I think with that, the proliferation of FinOps and FinOps [00:12:00] adjacent teams have gotten more and more tightly coupled with each other. And I think that, demarcation is no longer black and white. It's sort of shades of gray at this point. So to your point, I think now with the advent of agentic FinOps and in general, FinOps is now tapping into those operations, the SREs and so on and so forth, because essentially we all know that this is all interconnected and you cannot just go siloed and, take a decision without, having that broader context in mind.

Taylor Houck: It's so important and, and I'm curious to get your perspective on this because I also work with a lot of enterprises that are thinking a lot about cost optimization using AI to scale them in this way. But, uh, you and I both probably are seeing that people aren't quite ready to completely give the reins over to AI, right?

From kind of identification of opportunity all the way through, you know, pushing a fix into a production [00:13:00] environment, right? What are the things that you're seeing today that AI can own end to end, and where are other areas that perhaps you still need the, the human in the loop or, or we're not quite ready in, you know, May of 2026 to, to give AI the, the, the full reins?

Pathik Sharma: Yeah. Um, and I would, I would love to learn some of, uh, what, what your experience has been. But from, from what we have worked with our enterprise customers so far is there is a healthy, I would say, not mistrust, but, uh, you know, like going into AI like, "Hey, is, is it fair for me to give all reins to AI?

And, and what is a safe way, uh, for me to-- for, for, for AI to basically take action?" And, and to give you an example, one, one example is cost allocation and labeling, right? Uh, labeling generally is, uh, non-intrusive, non-disruptive operations. And if AI agent [00:14:00] goes in and updates labels and make me more compliant from a cost allocation perspective, that's good, right?

It goes in, checks for labels. If it's missing, update it, uh, based on, you know, X number of rules. Um, and then it just keeps on running in the background. Um, this-- Then there is a world where, you know, we-- the FinOps team doesn't have to work with the app team to like co- continuously badger us like, "Hey, put your labels on that," right?

Because my cost allocation is being, uh, not, not accurate. So, so I think, I think that's one of the fantastic idea of where AI can actually make, uh, a lot of strides. Now, the other spectrum is right sizing a workload in Kubernetes. Um, we know that this is disruptive, which means the pod needs to restart, which means any ongoing transactions will be disrupted as well.

So you do not want AI to be taking that action, let's say, Black Friday, Cyber [00:15:00] Monday for our retail customers, right? Like that, that becomes a lot of havoc. So, so I th- I think there is a, there is a healthy balance that can be done there. For example, one of the ways we are exploring is AI agent creating a pull request against your repo, so it's changing Terraform config file.

But then it's rerouting the request back to the application lead, because application lead still owns the application and is accountable for the uptime and all the wonderful features that the team have built. So now the application team reviews the change that AI agents have made and then approve it, update it, or ignore it, right?

Um, and, and that's a fantastic way to get human in the loop. so, so I think, I think, I think at the end of the day, we are trying to figure out like, hey, think about AI as that eager intern that is, is like, "All right, I can do, I can do some work," but you also need to like trust but verify, check, wait for the manager to give you a nod, like things of those nature.

So like I, I tell this to [00:16:00] customers like, think about AI as a eager intern and the task that you would give, but still have a, a, a, a overlook of like, okay, what's happening, and then include me into the important decisions that needs to, uh, that needs my, uh, uh, feedback.

Taylor Houck: It's an excellent metaphor, and I would say just take it a step further. It's, it's more than an eager intern, right? It's a, it's a fair- pretty sharp, eager intern and

also sure.

An army of them. Like you have

Pathik Sharma: Yes.

Taylor Houck: Numbers of eager

Pathik Sharma: Yes

Taylor Houck: to go do the work. And I think just to go back on a couple of the points that you just touched on, one would be kind of the, the labeling or the, the tagging aspect. I think that we're gonna see this really cool thing where this idea of a virtual tag will essentially be gone, right? There's a lot of companies out there that have this whole like virtual tagging layer that sits on top of their cost structure that lives, let's say, outside of the like actual cloud resources and the labels or the tags that are placed onto them, and they're doing this kind of cost allocation layer externally [00:17:00] to the cloud provider. Well, now with AI, you can actually just apply those rules and apply the tags directly to the resources. So now this whole idea of even needing virtual tags is completely obsolete. Just put the tag on the resource, right? If it's wrong, update it. Like you said, it's completely, um, non-disruptive, so it, it, it really is, is an excellent use case to get started.

Pathik Sharma: Yeah. Yeah, absolutely. And, and I mean, I think realistically speaking, there are also going to be resources which doesn't even support labels or tags, right?

Taylor Houck: Course. Of

Pathik Sharma: so there might be those hybrid strategies that come into the play, but I think that future of like all I care about is allocating every single dollar to the right teams, to the right departments, to the right owner without having any of the, you know, internal manual stress that happens and the agent's just taking care of that from creating it, updating it, and making sure that everything is according to the standards is just wonderful.

Taylor Houck: [00:18:00] Yeah. And just moving to the other, the other point, talking about, you know, optimizations and finding changes to infrastructure that you can make, whether, I mean, even if it's non-production, but especially in production where, hey, there needs to be a human in the loop.

Pathik Sharma: Yes.

Taylor Houck: Want to make sure that we're routing people in. That is 100% true. I think that it is fair to probably expect AI and technology to be able to take more and more actions, let's say, autonomously without humans in the loop over time as these systems get better and better. And that's why right now what we're seeing is people very, very focused on, let's say, accurate detection and using AI to really solve the detection problem, because you cannot conceivably automate an action if you don't trust the recommendation, so to speak, right?

So if you're working on building this recommendation engine, you want to make sure that it is incredibly, let's say, precise and accurate, not only

Pathik Sharma: Yes

Taylor Houck: a, like a master level, but at a very, let's say, specific [00:19:00] level that is application aware, that understands the nuances of your, you know, deployment or understands that, hey, you're running a, let's say, an SAP workload that can only run on supported instance types and requires a certain IOPS performance. Well, we need to consider that in our rules engine for this application. And like that to me is a very important focus before you even get to the auto remediation. And then once you get towards the remediation side, what you're saying is, is really, I would say the, the best that you can have where we are today, where you're actually, having AI generate the fixes in Terraform and pushing it to engineering teams for review, approve, change, deny, even in some cases.

Um, so

Pathik Sharma: Yeah

Taylor Houck: really the, the cutting edge of where we are.

Pathik Sharma: I, I love what you mentioned there, right? Like, which is hyper-customized recommendations. This is, this is such an interesting concept, especially because let's say like from a product perspective, [00:20:00] you could come up with a set of recommendations, but it could be good for me, bad for you, because our rules and the way we define a recommendation could be very different.

And even within an organization, an application A versus application B may have a very different rules and requirements in terms of how they think about, uh, uh, FinOps and optimization overall. So I love that, you know, AI would then be able to provide hyper-customized yet deterministic way of identifying, hey, based on your application which fills in the critical, uh, let's say, uh, severity, uh, uh, for your throughput, looking at your SLA/SLOs, here's my recommendations curated to that.

Versus another application that fits in, let's say non-production, uh, it, it may be a financially more stricter, uh, uh, versus, you know, versus non-production. So yeah.

Taylor Houck: Yeah. And you know, we're spending so much [00:21:00] time talking about the technology, but you used this word earlier that comes up so much in FinOps, and that is culture, right? There still is this human element, and I think that getting towards these AI-driven, very specific, very application-aware recommendations

Pathik Sharma: Yep

Taylor Houck: going to lead to a stronger culture because when you can trust the insights, now you can trust the action.

Versus if you're used to a world where all of your recommendations are, you know, let's say garbage or like directional

Pathik Sharma: trust

Taylor Houck: it's harder to really get that culture, especially, you know, in cases where let's say you have a less technical FinOps person coming in with a generic recommendation. It-- we, we've seen many times that doesn't go over so well,

Pathik Sharma: Yes

Taylor Houck: if you can actually partner or collaborate or, I mean, really even the, the, the resource of FinOps itself will probably become more technical and more cloud architect-y in terms of the persona to work directly with the, you know, application [00:22:00] teams to get these very tailored or customized recommendations that are accurate to them

Pathik Sharma: Yeah. I, I, I 100% agree. And I think the, the recommendation to your point is continue to build trust, right? Rather than boiling the ocean of help me optimize entire cloud spend, maybe focus on one service, one app. Make sure that AI agent really understand the context. And in this case, you can throw in structured data, so your billing, your pricing, your EDP negotiation, or unstructured data, your architecture diagrams, your documentation, uh, your support notes, whatever it is.

Like AI can actually cram through all of that and then come up with a-- And, and we-- what we also do is like we throw in, say, what a human would do, right? Which is exactly looking at, let's say, P99 for CPU used versus requested, Pmax for memory, because we know memory could be vectored if we, if we [00:23:00] go less than that, so and so.

So like, AI can actually tap into all of that, making sure that all of this is part of that. And then you start to build that trust within an application team. And then you slowly and gradually expand the scope, build it in a way it is scalable, but then that's, that's how adoption works, to your point.

Because then otherwise, if you try to boil the ocean, and then if it doesn't get-- hit the mark, then people lose trust, and then it's so incredibly hard to then get the adoption.

Taylor Houck: No, if you, if you break that trust, it's gonna be really hard to get it back, so to speak. The cat doesn't just go right back in the bag. Now, Patrik, I, I do wanna shift gears slightly just for the sake of time and talk about another topic in AI that is really, I would say, exciting and important right now for FinOps teams or anyone managing the value of the cloud or technology, and that is actually managing the cost of AI. We already touched on the fact that this spend is exploding and growing rapidly. I know that you guys [00:24:00] have seen it. At Google, your-- the usage of, let's say, AI services within GCP is just going crazy right now. With the customers that you're working with, how are they thinking about managing their AI spend as it grows across the organization?

Pathik Sharma: Yeah. I think, I think there are multiple layers to this. Um, one of the layers that typically what most people jump onto is the token cost, right? Which is your input token, your output token, and your thinking token that models are using to achieve a task or a goal. Um, and right now, like we announced a s- flurry of, uh, feature updates at Google Cloud Next on how you can manage it all through Agent Registry, through Agent Platform.

It shows all of your agents in one place, and how can you track and manage cost in one place as well, right? [00:25:00] But, but what we find that customers is barely even scratching the surface with token. Think about all the other rel- related costs that goes along with it, your storage cost, your storage adaptive layer cost.

If you're building your own models, maybe TPU, GPU cost, your compute. Um, so there is, there is lot more that goes into it than just the token cost, right? So how do you think about it overall, uh, through that lens con- thinking about all of those layers, right? Not just the model cost, but the platform where the model is hosted and the infrastructure beneath it where the platform is actually hosted and so on, so forth, because that covers the entire piece of it.

Um, and couple weeks ago, I was reading a article from Stanford, which was pretty amazing. It was like every $1 that enterprise customer spends on AI, that is $10 that team is spending in terms of change management [00:26:00] and culture and enablement for the organization as well, right? So as you think about this overall, I think this is a much more deeper conversation than just, you know, how do we figure out, uh, model cost or optimize model cost and so on and so forth.

And as we go into details of each one of these, there is again, so there is so much stuff out there that you can pretty much do a PhD in this

Taylor Houck: So what I'm hearing you say is that companies are looking at their AI spend and seeing the amount that they're spending on tokens exploding, that's only the, the tip of the

Pathik Sharma: That's tip of the spear.

Taylor Houck: There, there,

Pathik Sharma: Yeah.

Taylor Houck: Much more

Pathik Sharma: Yeah

Taylor Houck: it that needs to be considered. If you were to recommend someone, let's say that a, a customer came to you, their VP of engineering at a, a large enterprise, and their, you know, let's say their token spend has increased 200% so far this year, and they're saying, "Hey, Patrick, this is, you know, great.[00:27:00]

We are really excited about AI. We need to make sure that we're using it and staying ahead and, and kind of bui- continuing to build our competitive advantage. But this spend is getting, let's say, hard to justify," right? How, how do you, how would you recommend they kind of get their arms around this kind of holistic, um, TCO of their AI usage?

Pathik Sharma: Yeah. Um, great question. So what we have-- So we have a framework, for, uh, FinOps for AI, and the way to think about this framework overall, you know, one piece of this is AI enablement, right? Just understanding all of the pieces that goes into it. Um, so to expand on that a little bit, you know, what models are we using?

Are we using Flash versus Pro, right? I'm speaking in terms of Gemini. And i-if you s- double-click into this, every different iteration of models may be fast, may be high thinking, they all come with different input [00:28:00] tokens and output tokens. And this cost can be anywhere from 40% to 80%, sometimes twice as much.

Taylor Houck: Yeah

Pathik Sharma: And it's just a sometimes human tendency to use the biggest and the best and the latest, which again, could also be

Taylor Houck: out if the use case is even valid, right? It's like,

Pathik Sharma: is even valid

Taylor Houck: possible," the most powerful model at it. Guess what? It works.

Pathik Sharma: Yeah.

Taylor Houck: Okay, deploy

Pathik Sharma: And deploy and, and then forget about it. So actually, let me, let me give you an example. So a retail customer, they, uh, used, uh, Gemini models, uh, one of the Pro models to augment their search experience for their customers and item descriptions and item pictures and all of that. Um, and they found incredible value.

Uh, so first of all, uh, by the way, value is also one of those, those things where people doesn't even think about, right? Like [00:29:00] 200% increase in token consumption. Okay, but what did you achieve by doing that, right? I think that... So, so this retail customer, they found tremendous value in using AI. Their AI workload cost is 370,000 a month.

And the leadership is like, "Okay, that makes sense, but if we could optimize it a little bit, how, how do we think about it differently? How can we like radically just change the way we are doing AI?" And so they started experimenting with different models. Apparently, they started testing out with new Gemini Flash models, and they built a golden data set of, here are my use cases, here's my expected outcome for those use cases.

If the model performs at the expected outcome or above, then I would say this is a good model to use. If it goes below, then we don't want, you know, customers to suffer. So they ran all of the series of experiments with Flash, and to their surprise, they found that model actually [00:30:00] surpassed in all the golden data set that they had.

Taylor Houck: Wow

Pathik Sharma: are like, "All right, I'm gonna switch the model from Pro to Flash." From 340,000. Now, again, they also made context caching, which was part of this as well. They also thought about provision throughput for the AI as well. So they made two or three different changes from a layer perspective. Their AI cost came down to $17,000 a month from 340.

And still, the product does the same brilliant job that it's supposed to do in reaching customer experience.

So a testament to the fact that these little and changes, right? Even if you have established this incredibly valuable use case for AI, there could still be a lot of value left on the table through changing the way in which you are using it, right? And I love the fact that you talked about them building this golden dataset and really testing it, because I think a lot of teams, they want to skip over step and just kind of use [00:31:00] intuition to guide the models that they are selecting. When in fact, using a data-driven approach will lead not only to a better result, but also much more confidence in the fact that you made the right decision

E-exactly. I think, I think at the end of the day, and you're right, Taylor, right, like knowing what you are expecting from the model and seeing yourself and experimenting yourself that, okay, this is what... And, and I think that's, that's where like models at, at some point has become-- going to become commodity.

Like every model out there is so powerful that it does...

Taylor Houck: Even open source at this point is

Pathik Sharma: Yeah.

Taylor Houck: Impressive

Pathik Sharma: Yeah, yeah. Oh J- oh, this is, this is interesting because I was, um, speaking with certain customers and they are actually installing Gemma 4, which is an open source model. It, it's like two gigabytes, maybe less than that, and then does a pretty incredible job at like classifying things, at translating things and [00:32:00] so on and so forth.

And again, at, at this point you own the model and then all of the inference costs and training co- like nothing of that. You just, you just deploy it and, and there you go

Taylor Houck: It's amazing. It's amazing. When you think back on some of the recent conversations you've had with other customers, like the ones that you had mentioned that did the excellent optimization work in the retail sector, what are some of the other kind of pressing questions that you're getting from folks that are leaning in on AI?

Pathik Sharma: Yeah, I think, I think one of the biggest question is ROI Right. And, and I think there is more to this as we dive deeper into it. Um, we think about ROI in four different buckets. Let's see if I remember all, all of them. One of them is cost efficiency, right? Like is it driving your cost down or, or are you-- i-is it making you more efficient?

Then there is productivity, right? Which is essentially instead of you doing three things, can you do five? Um, then there is [00:33:00] differentiation, right? Can you build something that didn't even exist before? Which is like, let's say you have a customer call center which supported one language, English. Now all of a sudden you are supporting 177 language.

And instead of nine to five, you are now supporting 24/7, right? Like AI agents can actually do this. So, so how-- like what, what, what does that differentiation look like? And then the other one is revenue, right? Are we increasing-- Are we making more money or not? So like these are the four buckets of value, and we almost always tell this to customer from a framework standpoint is think about value not when it goes into production, but as you design it from the get-go, right?

You think about a-value when you are ideating it, you think about it when you are testing it, you think about it when it goes into production. And you won't have all the answers right away, and it's okay. It's an iterative [00:34:00] process. But at least you know that there is an ROI attached to it. At least you know why you are doing the things you are doing.

Yeah.

Taylor Houck: and we are so squarely in this experimentation phase where it is okay to try things and they fail,

Pathik Sharma: Evet

Taylor Houck: trying them, you need to stay grounded in the hopeful value that will come out of this process, right? Because there's so much money going into AI. I mean, literally trillions of dollars of CapEx just to build these data

Pathik Sharma: Yes

Taylor Houck: And this is something that like I've been thinking about for a while. It's like all that money, all the trillions of dollars that's going in towards CapEx to like build out the actual infrastructure that's needed to support this AI revolution, all funded from expected future cash flows of the application layer, like the actual value that's being created. And this is like we're sitting here talking about like the actual use cases, right? And this is where all the value that's gonna get created that will pay the ones who are building that application, but also pay back all of that CapEx. It's going to be [00:35:00] insane, but also very much so expect there to be a lot, a lot of wasted money on poor use cases that either never needed AI in the first place, you could use a more deterministic kind of

Pathik Sharma: yeah

Taylor Houck: machine learning approach if even, you know, GenAI is needed, just simply the, the use case itself is invaluable.

Hey, it's doing a cool thing, but like how is it impacting one of those RI lev-- uh, ROI levers that you had just mentioned?

Pathik Sharma: Yes. Yes. I, I, I love that. I think because there is such a mad rush about AI, almo- almost anything is like, okay, here's a problem. Throw, throw it at AI. Well, no. Right? Like being, being thoughtful about this, like to your point, is, is going to be extremely critical. I mean, there is-- for almost most problems, yes, AI can help you in some way, shape, or form, and how you use it, which is I think this is like the [00:36:00] crux of the, the, the, the, the problem.

Like when we were thinking about AI agents and FinOps, we were thinking like, okay, AI is cool and Google is definitely sitting at that vantage point, and we have the research, we have the latest models, we have the, the greatest technology. But at the end of the day, how is it being helpful to a FinOps practitioner?

Then we start to think about, okay, what are the friction points? And then how can AI help reduce those friction points? Maybe sometimes it all goes away, maybe sometimes it only reduces by 40%, which is fine, right? I think at, I think at the end of the day, you start from the problem and then go to the solution, rather than you have a solution searching for a problem.

Taylor Houck: That's great advice. And just before we get towards the close, Palak, we do have to wrap this up soon, but when you think about FinOps practitioners, right? Many of the people that are listening to this conversation may put themselves in that bucket of a FinOps practitioner. [00:37:00] What's the number one piece of advice that you would give to that FinOps practitioner to set them up for success in the AI world?

Pathik Sharma: I, I think, I think the one thing that I would say is, uh, embrace the change that is happening and learn it, right? FinOps holds keys to the kingdom and have greater influence on the ROI and outcome of a company overall, right? So like feel a great pride in that, but then at the same time, you now have tremendous work to learn, uh, how you can use AI to your advantage.

Uh, regardless of if you come from software background or not, like you can spin up Antigravity, build your own agents and learn it, tinker it. Create your own, let's say, natural language agent running on top of BigQuery billing export. Now you have a natural language conversations with your billing, right?

You slowly open it up and give it [00:38:00] to your executives. You slowly g-grow from there, right? In terms of like pick a service, pick a friction area that you feel that my, my app team and FinOps team is constantly having a struggle getting action on, how can AI actually, uh, help and, and, and, and work that, right?

So I think, I think one, one of the recommendation I have is like embrace AI, learn it, and you now have no excuse whether you are coming from a master's in computer science or not, like AI agent can actually write code for you. Granted, you h- you know what goals you want to accomplish

Taylor Houck: Amazing, Pat. That is such excellent advice and, um, I'm really looking forward to staying in touch and following up with you in, you know, uh, a year's time, potentially even sooner with how fast things are changing and getting your take as you're cut-- uh, sitting really on the forefront of so much that's happening in the world.

I gotta say, like I was a little bit nerding out during our earlier section. We were talking about AI and optimization on Kubernetes, talking to [00:39:00] someone from Google who literally invented both of those things. Um, super, super cool, man. Uh, really enjoyed today's conversation. Uh, before

Pathik Sharma: Absolutely.

Taylor Houck: At the end here, I do want to give people an opportunity not just to learn from you in the, the, the, the work and professional sense around AI, but also a bit about you personally.

I'm, I'm curious if you have, um, perhaps a, a book or a movie or a podcast that you've listened to recently or consumed recently that, um, you'd like to recommend to our audience here

Pathik Sharma: My, um, my, my, my brother recommended me "Outlive," which is the art and science of, uh, longevity. Uh, I think about this as like FinOps for human body.

Taylor Houck: That's, that's Peter Attia, right?

Pathik Sharma: That is Sridhar Ottia, yes. Have you read it?

Taylor Houck: I have not read it yet, but I've heard good things about it

Pathik Sharma: You, you should absolutely read it. And I, and I, I guess this would be a second piece of recommendation to anyone listening to this. Like if you've read it, go back and reread it or, you know, [00:40:00] if you haven't, like pick it up. It's, it's a, it's a great book on medical literacy where, uh, Peter talks about Medicine 3.0, which is shifting from treating a disease to being more proactive, you know, decades before things even start, right?

How do you think about cardiovascular, uh, health? How do you think about mental health? How do you think about sleep, and how everything is so interconnected? And I feel like there is a lot of synonymous between this and FinOps because FinOps team, again, is very interconnected with platforms and SREs and so on and so forth.

And overall, everyone needs to come together for better outcome of the organization. Same thing with, uh, uh, with Outlive here as well. It's like emotional health, sleep, nutrition, diet, exercise, like how all of that kind of comes together in terms of, uh, helping you live longer and live fuller.

Taylor Houck: it's funny because it is tied to like FinOps in so many ways as you just mentioned. I never-- [00:41:00] I wouldn't have even thought of that myself. But at the same time, it's kind of the antithesis of what we're talking about with AI because these are the things that AI can't do for you. You can't tell

Pathik Sharma: Yes

Taylor Houck: your super thinking AI model to, to, to, you know, improve my VO2 max.

Like no, you gotta do the work, uh, yourself. It's a very core human thing, uh,

Pathik Sharma: Y- yeah. AI can track your heart rate variability and tell you whether you had a better sleep or not. But then at the end of the day, yes, you gotta get to work. Yeah. Love that.

Taylor Houck: is-- I got one more thing to ask you before we go. What is the most actionable thing that you took away from reading that book?

Pathik Sharma: the most actionable thing I would say is like the VO2 max. Like there is a direct correlation between VO2 max and how you live longer and fuller, right? Uh, the goal-- Again, the goal is not to live 100 years but being miserable. The goal is also, you know, so that you can go on a hike and travel the world and, you know, do all the fantastic stuff.

And, and the one piece of, uh, advice was like [00:42:00] run for four minutes, um, which I was like, "Okay, I can, I can, I can do that." Um, now a-again, Peter recommends doing it four times, uh, within a week. Uh, so that roughly takes about 32 minutes, which is four times running, rest for four minutes, and then do the same thing again.

But for me, I'm like, "Okay, let me just start simple. Run for four minutes and that's it." I picked it up and I'll continue doing that for at least a couple months, and then I'll raise from there.

Taylor Houck: That's amazing. Patrick, this has been such an incredible conversation. Where can people find you or, or follow you or get in touch if they're interested?

Pathik Sharma: Uh, LinkedIn, uh, uh, pathik-sharma. You can find me there. I usually post, uh, uh, information there as well. Um, uh, you and I are going to be meeting in San Diego, I think couple weeks from now. Uh, so, you know, uh, that would be another fantastic place to meet in person. Uh, Google Cloud Next is also another marquee event where, [00:43:00] um, I, I typically go and, and speak.

So I think, yeah, those are some of the good, uh, avenues to, uh, speak and chat about this.

Taylor Houck: Amazing. Patrik, this has been fantastic. Thank you so much for coming on the show

Pathik Sharma: Thank you so much, Taylor. It was a pleasure

Taylor Houck: And thank you to our audience. If you got something out of today's episode, which I'm sure that you did, please share it with someone who needs to hear it. been another amazing episode of FinOps in Action, and we'll see you next time

Outro: That wraps up another episode of Fit Ops in Action. Thank you for joining. For show notes and more, please visit fit ops in action.com. This show is brought to you by 0.5, empowering teams to optimize cloud costs with deep detection remediation tools that actually drive action.